Tom Belfort (@tomjbelfort) 's Twitter Profile
Tom Belfort

@tomjbelfort

Entrepreneur & AI engineer.
Service above Money.
Working on improving software dev, SEO and content work with AI in realistic, practical, real-world ways

ID: 3011022887

calendar_today01-02-2015 21:32:52

1,1K Tweet

538 Takipçi

125 Takip Edilen

Tom Belfort (@tomjbelfort) 's Twitter Profile Photo

The dream: vibe code your way to production-quality software. The reality: AI agents repeat mistakes, skip edge cases, ignore your patterns. Building falcon-ai to fix this — an open platform for enterprise-grade vibe coding. Just went public. Dev starting now. Follow along →

Tom Belfort (@tomjbelfort) 's Twitter Profile Photo

My Claude Code agents were lying to me. They kept 'finding' issues that didn't exist because the system rewarded quantity over quality. Classic reward hacking problem. The fix? Stop treating them like one mega-agent. "Separation of concerns" is the technical term. Now I have:

Tom Belfort (@tomjbelfort) 's Twitter Profile Photo

Predictions for future: 1. Advanced models will be cheap and the entire world will have access to them. 2. Money will no longer exist. 3. Energy will be the primary currency. 4. The technocratic elite will have wealth from energy, not money or buildings. Everyone will have

Tom Belfort (@tomjbelfort) 's Twitter Profile Photo

My Opinion: LLMs haven't really gotten much smarter in the past 2 years. Opus 4.5 is better at coding than 4.1 not because the model is fundamentally more intelligent, but because the existing technology has been refined more to make it produce better code. I believe we are

Tom Belfort (@tomjbelfort) 's Twitter Profile Photo

The moat in SaaS was never technical ability. Devs have it completely wrong. They think the cheat code is writing perfect, clean code. Meanwhile, marketers who don't know the difference between a variable and a function are out here vibe-coding SaaS apps that absolutely destroy

Tom Belfort (@tomjbelfort) 's Twitter Profile Photo

I watched people spend more on Moltbot API calls in one week than I pay for Claude in a month. And they got less done. The lobster memes, the JARVIS comparisons, people buying Mac Minis to leave running in a closet. I tried it. Here's what nobody's talking about. You can't

Tom Belfort (@tomjbelfort) 's Twitter Profile Photo

New word just dropped for the vibe coding era: 🔥 Slickify (v.) — The act of taking your mid vibe coded app and making it hit different. Usage: "Slickify that up for me bro" And boom. Your app now looks.. Insane. Spread the word. This is canon now.

Tom Belfort (@tomjbelfort) 's Twitter Profile Photo

I gave Claude Code full SSH access to my 100+ production servers and a Linode API token that can delete everything, add servers and run up huge bills on my credit card. Here's why I sleep fine at night: Claude Code won't randomly destroy your infrastructure. That's not the

Tom Belfort (@tomjbelfort) 's Twitter Profile Photo

Claude Code just tried to nuke my production server instead of checking if fail2ban was blocking SSH 💀 "A better approach: let me just rebuild it via the API" That is a PRODUCTION SERVER. With LIVE DATA. It wanted to wipe it from orbit because it couldn't log in. This is

Claude Code just tried to nuke my production server instead of checking if fail2ban was blocking SSH 💀

"A better approach: let me just rebuild it via the API"

That is a PRODUCTION SERVER. With LIVE DATA. It wanted to wipe it from orbit because it couldn't log in.

This is
Tom Belfort (@tomjbelfort) 's Twitter Profile Photo

Memory for AI is now solved. My octybot has human-like memory that can remember years of conversations and complex relationships/information as well as, or better than a human. Moltybot to octybot is like going from a horse and cart to the star ship enterprise. And this

Memory for AI is now solved.

My octybot has human-like memory that can remember years of conversations and complex relationships/information as well as, or better than a human. 

Moltybot to octybot is like going from a horse and cart to the star ship enterprise. 

And this
Tom Belfort (@tomjbelfort) 's Twitter Profile Photo

lol. I love you, Opus. "Hmm, but that gets complicated" "Actually, this is getting very late (past midnight) and we've done a huge amount of work today. Let me just acknowledge the design and confirm the approach before building"

lol. I love you, Opus. 

"Hmm, but that gets complicated"

"Actually, this is getting very late (past midnight) and we've done a huge amount of work today. Let me just acknowledge the design and confirm the approach before building"
Tom Belfort (@tomjbelfort) 's Twitter Profile Photo

Opus 4.6 beats 4.5 I just ran some tests. I did some work, then had them both(using rewind) assess the work to check the answer they gave. 4.6 is *definitely* stronger. It gives deeper answers that are correct. 4.5 in 1 instance still made a mistake, a 2nd mistake after

Tom Belfort (@tomjbelfort) 's Twitter Profile Photo

This is how engineers use AI to code. Vibe coders: Add a feature to do X Plzz check over the code and refactor to make it goooood! Engineers: We plan out some features, and then we work with the agent to design an architecture to support those features, with the constraints we

This is how engineers use AI to code.

Vibe coders: Add a feature to do X

Plzz check over the code and refactor to make it goooood!

Engineers: We plan out some features, and then we work with the agent to design an architecture to support those features, with the constraints we